In this episode of Product Craft, Arseni spoke with TJ Pitre, founder of South Left Agency.
TJ has been working in design systems for over 15 years, and more recently, helping teams figure out how AI actually fits into real product workflows—not just in theory.
We talked about where things stand today, what’s actually working, and where teams are still struggling. No hype—just what it looks like on the ground.
What you’ll get from this episode
A real-world picture of where AI adoption is
What’s working today vs what’s still messy
Where design systems break in real product environments
How designers can start working closer to production
Why context matters more than prompting
Key takeaways
1. Most teams are still figuring AI out
There’s a big gap between awareness and real usage.
Some teams are going all in. Most aren’t (yet). What TJ sees across clients is a middle ground—teams experimenting, setting up internal AI groups, trying things out, but not fully integrating it yet.
So if things feel unclear—that’s normal.
2. The real gap isn’t design → dev
Everyone talks about the handoff between design and engineering.
But the bigger issue shows up later—when components actually get used in the product.
That’s where things start to drift:
deadlines kick in
edge cases appear
engineers make adjustments to ship
And suddenly, the “system” isn’t so consistent anymore.
3. Prompting isn’t the hard part
Early on, people focused on prompt engineering.
In practice, that’s not the bottleneck.
What matters more is how much context the system has:
components
tokens
patterns
constraints
If that’s messy, the output is messy. If it’s clean, things start to work.
4. Design systems are becoming more than libraries
Most design systems today are still static.
What’s emerging is something closer to a system that can actually respond to context—generate UI, adapt layouts, and plug into workflows more directly.
It’s early, but you can see where this is going.
5. Figma still matters, but differently
You can build products without touching Figma now.
But it’s still useful—just not in the same way.
More as:
a place to explore ideas
a way to think through UI
a bridge when something is easier to see than to code
In practice, it’s less about “Figma vs code” and more about using both where they make sense.
6. Designers are getting closer to the product
This is probably the biggest shift.
With the right tools, designers can:
generate components
test interactions
work directly with real UI
Not perfectly—but enough to meaningfully contribute.
That changes the role quite a bit.
7. Most design systems look fine—but aren’t
One line from TJ stuck with me.
Most design systems are like a car with the check engine light on.
Everything works. But under the hood:
things aren’t named properly
tokens aren’t consistent
components aren’t fully thought through
You don’t notice until you try to scale—or plug AI into it.
Why this matters
The tools are moving fast.
But the real constraint isn’t tools—it’s how systems are set up.
The teams that benefit from AI won’t be the ones trying random tools.
They’ll be the ones with:
clear structure
clean systems
and a good sense of how things actually get built
Resources & Links
TJ Pitre (guest)
TJ on Linkedin: https://www.linkedin.com/in/tpitre/
Southleft: https://southleft.com
Figma Lint: https://www.figma.com/community/plugin/1521241390290871981/figmalint
Sequential Thinking: https://github.com/modelcontextprotocol/servers/tree/main/src/sequentialthinking
Serena: https://github.com/oraios/serena
Design System Assistant MCP: https://design-systems-mcp.southleft.com
A2UI: https://a2ui.org/
Arseni Harkunou (host)
Linkedin: https://www.linkedin.com/in/arseni
STILL HUMAN: www.stillhuman.com
Personal website: www.arseni.com
